Closed weberhen closed 7 years ago
Ho god indeed I thought that this bug was already fixed, actually, we sould find a way to just bring the tensor to cpu, and not convert the type, it is not train loop's job to determine those types. I think you can call cpu() method instead. Make a pull request with those fix if it works!
Fixed here #18
When I try to run on CPU mode I have a problem:
1) on "setup_loaded_data" it will set the data to float and target to long (line 64) , which gives me a torch error of mismatching types in the loss function. So my workaround is convert target to float and not to long.